A New Asymmetric Link-Based Binary Regression Model to Detect Parkinson's Disease by Using Replicated Voice Recordings

摘要

Addressing dependent data as independent has become usual for Parkinson's Disease (PD) detection by using features extracted from replicated voice recordings. A binary regression model with an Asymmetric Student t (AST) distribution as link function has been developed in a classification context by taking into account the within-subject dependence. This opens the possibility of handling situations in which the probabilities of the binary response approach 0 and 1 at different rates. The computational issue has been addressed by proposing and using a representation based on a mixture of normal distributions for the AST distribution. This allows to include latent variables to derive a Gibbs sampling algorithm that is used to generate samples from the posterior distribution. The applicability of the proposed approach has been tested with a simulation-based experiment and has been applied to a real dataset for PD detection.